Skip to content

Consolidate possible optimizations #31

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jrood-nrel wants to merge 3 commits into
base: main
Choose a base branch
from
Open

Consolidate possible optimizations #31

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
82eefce
Try a few optimizations.
jrood-nrel May 30, 2024
b7d3947
Add assert.
jrood-nrel May 31, 2024
13d3eea
Fix shadowed variables.
jrood-nrel Jun 3, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions src/exchangeDonors.C
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -35,7 +35,6 @@ void tioga::exchangeDonors(void)
// and receiving
//
pc->getMap(&nsend,&nrecv,&sndMap,&rcvMap);
if (nsend == 0) return;
//
// create packets to send and receive
// and initialize them to zero
Expand Down Expand Up @@ -74,7 +73,8 @@ void tioga::exchangeDonors(void)
//
// communicate donors (comm1)
//
pc->sendRecvPackets(sndPack,rcvPack);
pc->sendRecvPackets2(sndPack,rcvPack);
if (nsend == 0) return;
// Initialize linked lists and populate donor data from rcvPack
for (int ib=0;ib<nblocks;ib++) {
auto& mb = mblocks[ib];
Expand Down
133 changes: 133 additions & 0 deletions src/parallelComm.C
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -212,6 +212,139 @@ void parallelComm::sendRecvPacketsAll(PACKET *sndPack, PACKET *rcvPack)
// TIOGA_FREE(status);
// }

void parallelComm::sendRecvPackets2(PACKET *sndPack,PACKET *rcvPack)
{
int i;
int *sint,*sreal,*rint,*rreal;
//
sint=(int *)malloc(sizeof(int)*numprocs);
sreal=(int *) malloc(sizeof(int)*numprocs);
rint=(int *)malloc(sizeof(int)*numprocs);
rreal=(int *) malloc(sizeof(int)*numprocs);
// remove when using stl vectors and just init the vectors to 0
for(i=0;i<numprocs;i++){
sint[i]=sreal[i]=0;
rint[i]=rreal[i]=0;
}
for(i=0;i<nsend;i++){
sint[sndMap[i]]=sndPack[i].nints;
sreal[sndMap[i]]=sndPack[i].nreals;
}
//
MPI_Alltoall(sint,1,MPI_INT,rint,1,MPI_INT,scomm);
MPI_Alltoall(sreal,1,MPI_INT,rreal,1,MPI_INT,scomm);
//
for(i=0;i<nrecv;i++) {
rcvPack[i].nints=rint[rcvMap[i]];
rcvPack[i].nreals=rreal[rcvMap[i]];
}

int all_snd_nints = std::accumulate(sint, sint + numprocs, 0);
int all_rcv_nints = std::accumulate(rint, rint + numprocs, 0);
int *all_snd_intData, *all_rcv_intData;
all_snd_intData=(int *) malloc(sizeof(int)*all_snd_nints);
all_rcv_intData=(int *) malloc(sizeof(int)*all_rcv_nints);
for (int i = 0; i < all_snd_nints; i++) {
Comment thread Fixed
all_snd_intData[i] = 0;
}
for (int i = 0; i < all_rcv_nints; i++) {
Comment thread Fixed
all_rcv_intData[i] = 0;
}
std::vector<int> snd_int_displs(numprocs+1, 0);
std::vector<int> rcv_int_displs(numprocs+1, 0);
for (int i=1; i <= numprocs; i++) {
Comment thread Fixed
snd_int_displs[i] = snd_int_displs[i-1] + sint[i-1];
rcv_int_displs[i] = rcv_int_displs[i-1] + rint[i-1];
}
for (int i=0; i < nsend; i++) {
Comment thread Fixed
int displ = snd_int_displs[sndMap[i]];
for(int j=0; j < sint[sndMap[i]]; j++){
all_snd_intData[displ+j] = sndPack[i].intData[j];
}
}
MPI_Request int_request;
MPI_Ialltoallv(all_snd_intData,
sint,
snd_int_displs.data(),
MPI_INT,
all_rcv_intData,
rint,
rcv_int_displs.data(),
MPI_INT,
scomm,
&int_request);

int all_snd_nreals = std::accumulate(sreal, sreal + numprocs, 0);
int all_rcv_nreals = std::accumulate(rreal, rreal + numprocs, 0);
REAL *all_snd_realData, *all_rcv_realData;
all_snd_realData=(REAL *) malloc(sizeof(REAL)*all_snd_nreals);
all_rcv_realData=(REAL *) malloc(sizeof(REAL)*all_rcv_nreals);
for (int i = 0; i < all_snd_nreals; i++) {
Comment thread Fixed
all_snd_realData[i] = 0;
}
for (int i = 0; i < all_rcv_nreals; i++) {
Comment thread Fixed
all_rcv_realData[i] = 0;
}
std::vector<int> snd_real_displs(numprocs+1, 0);
std::vector<int> rcv_real_displs(numprocs+1, 0);
for (int i=1; i <= numprocs; i++) {
Comment thread Fixed
snd_real_displs[i] = snd_real_displs[i-1] + sreal[i-1];
rcv_real_displs[i] = rcv_real_displs[i-1] + rreal[i-1];
}
for (int i=0; i < nsend; i++) {
Comment thread Fixed
int displ = snd_real_displs[sndMap[i]];
for(int j=0; j < sreal[sndMap[i]]; j++){
all_snd_realData[displ+j] = sndPack[i].realData[j];
}
}
MPI_Request real_request;
MPI_Ialltoallv(all_snd_realData,
sreal,
snd_real_displs.data(),
MPI_DOUBLE,
all_rcv_realData,
rreal,
rcv_real_displs.data(),
MPI_DOUBLE,
scomm,
&real_request);

// FIXME: here and above I think I should move this a bit lower
Comment thread Fixed

Check notice

Code scanning / CodeQL

FIXME comment

FIXME comment: here and above I think I should move this a bit lower
MPI_Wait(&int_request, MPI_STATUS_IGNORE);
for(i=0;i<nrecv;i++){
if (rcvPack[i].nints > 0) {
rcvPack[i].intData=(int *) malloc(sizeof(int)*rcvPack[i].nints);
}
if (rcvPack[i].nreals > 0) {
rcvPack[i].realData=(REAL *) malloc(sizeof(REAL)*rcvPack[i].nreals);
}
}

// FIXME: here and above I think I should move this a bit lower

Check notice

Code scanning / CodeQL

FIXME comment

FIXME comment: here and above I think I should move this a bit lower
MPI_Wait(&real_request, MPI_STATUS_IGNORE);
for (int i=0; i < nrecv; i++) {
Comment thread Fixed
int displ = rcv_int_displs[rcvMap[i]];
for(int j=0; j < rint[rcvMap[i]]; j++){
rcvPack[i].intData[j] = all_rcv_intData[displ+j];
}
}
for (int i=0; i < nrecv; i++) {
Comment thread Fixed
int displ = rcv_real_displs[rcvMap[i]];
for(int j=0; j < rreal[rcvMap[i]]; j++){
rcvPack[i].realData[j] = all_rcv_realData[displ+j];
}
}

TIOGA_FREE(all_snd_intData);
TIOGA_FREE(all_rcv_intData);
TIOGA_FREE(all_snd_realData);
TIOGA_FREE(all_rcv_realData);
TIOGA_FREE(sint);
TIOGA_FREE(sreal);
TIOGA_FREE(rint);
TIOGA_FREE(rreal);
}

void parallelComm::sendRecvPackets(PACKET *sndPack,PACKET *rcvPack)
{
int i;
Expand Down
2 changes: 2 additions & 0 deletions src/parallelComm.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,6 +58,8 @@ class parallelComm {

void sendRecvPacketsAll(PACKET *sndPack, PACKET *rcvPack);

void sendRecvPackets2(PACKET *sndPack, PACKET *rcvPack);

void sendRecvPackets(PACKET *sndPack, PACKET *rcvPack);

void sendRecvPacketsCheck(PACKET *sndPack, PACKET *rcvPack);
Expand Down
3 changes: 2 additions & 1 deletion src/tioga.C
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -492,7 +492,8 @@ void tioga::performConnectivity(void)
this->myTimer("tioga::exchangeDonors",1);
//this->reduce_fringes();
//outputStatistics();
MPI_Allreduce(&ihigh,&ihighGlobal,1,MPI_INT,MPI_MAX,scomm);
//MPI_Allreduce(&ihigh,&ihighGlobal,1,MPI_INT,MPI_MAX,scomm);
ihighGlobal = 0;
Comment thread
marchdf marked this conversation as resolved.
//if (ihighGlobal) {
this->myTimer("tioga::getCellIblanks",0);
for (int ib=0;ib<nblocks;ib++) {
Expand Down